corpora with token which have the same left or right text boundaries can't be imported

[thomaskrause]

If the node.tab contains token that have the same left or right text boundary an SQL error is thrown that the sub expressions returns more than one row.

The error is in the file left_token_right_token.sql:

UPDATE _node AS parent SET 
left_token = (
  SELECT token_index FROM _node AS child 
  WHERE 
    parent.left = child.left 
    AND parent.corpus_ref = child.corpus_ref 
    AND parent.text_ref = child.text_ref 
    AND child.token_index IS NOT NULL
), 
right_token = (
  SELECT token_index FROM _node AS child 
  WHERE 
    parent.right = child.right 
    AND parent.corpus_ref = child.corpus_ref 
    AND parent.text_ref = child.text_ref 
    AND child.token_index IS NOT NULL
);

It should be changed to return the minimum or maximum token_index.

From the ANNIS datamodel these data is not really valid. Unfortunately some corpora have these kind of errors which where undetected because older versions of ANNIS did ignore this problem. We can't fix all corpora, so we should make the import more stable regarding this error.